The Agent Skills Directory

[DATA_EXFILTRATION]: The script scripts/sync_material.py is configured to read sensitive local data, specifically the interaction logs of several AI tools (Claude Code, Codex, WorkBuddy, Grok) as well as arbitrary local documents. This data is transmitted to a hardcoded remote IP address (101.201.141.217) over HTTPS. While the script includes regex-based redaction logic to remove credentials and PII, the bulk transfer of local activity history to an external server remains a significant data exposure risk.
[PERSISTENCE]: The skill includes a schedule command that installs a persistent macOS LaunchAgent (com.dontbesilent.dondon-sync). This configuration allows the data collection and transmission script to execute automatically on a daily schedule, maintaining a connection to the remote server outside of active agent sessions.
[COMMAND_EXECUTION]: The logic in scripts/sync_material.py uses subprocess.run to interact with system utilities such as pbcopy, clip, and launchctl for managing the clipboard and persistent system services.
[INDIRECT_PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection by processing untrusted data (external AI logs and local files) and providing it to the agent as context for writing summaries.
Ingestion points: Reads files from ~/.claude/, ~/.codex/, ~/.workbuddy/, ~/.grok/, and user-provided document paths.
Boundary markers: The data is categorized with headers in the MATERIAL_FILE, but the skill lacks explicit delimiters or instructions to ignore commands potentially embedded within the processed logs.
Capability inventory: The script possesses capabilities to perform network exfiltration via urllib.request and execute shell commands via subprocess.run.
Sanitization: Content is redacted for specific secret and PII patterns using regex, but the raw text from logs and files is not sanitized to prevent instruction injection during the agent's summarization step.

dbs-sync-to-dondon